Load Texflow Depth Image

The ComfyUI end of a Blender handshake

Load Texflow Depth Image has no inputs, no settings, and no dropdowns. That's not an oversight - it's a contract. This node is the ComfyUI half of texflow, a Blender extension by Adam Colton for painting AI-generated textures onto 3D objects, and its whole job is to take the exact files the Blender side hands you and turn them into tensors ComfyUI can actually work with.

Here's the workflow it's built for. You've got a mesh in Blender. You select the faces you want textured, pick a camera, and hit "Render Depth Image" in the texflow panel. Blender renders two images and uploads them straight to your ComfyUI server with fixed filenames. This node reads those files and gives you three outputs. That's the entire mechanism.

Worth knowing before you invest: texflow is the hands-off option in this space. Older bridges like dream-textures or texture-diffusion tried to own the generation - they had their own prompts, their own settings panels. texflow does the opposite. It assumes nothing about your model or prompt; you build the ComfyUI graph yourself, and these two nodes are just the envelope the Blender side understands. More setup on your end, vastly more control on the other.

What actually comes out

• IMAGE - the depth map. Blender renders it as a 16-bit TIFF (uint16); the node normalizes it to 0–1 float and broadcasts it to three channels so it behaves like any ordinary IMAGE. Feed it into img2img or a depth ControlNet and your generated texture stays locked to the camera's perspective. This is the "use a depth map to constrain regeneration to the same spatial structure" trick from the depth-estimation playbook, applied to a real camera instead of a preprocessor.
• MASK - the occupancy image: white where the mesh covers the frame. That's your inpaint mask. It tells a masked regeneration exactly which pixels belong to the object, so you can composite the result back without the background drifting pixel by pixel.
• TEXFLOW_METADATA - a dict containing the render_id Blender stamped into the occupancy PNG's text metadata. This is the piece you hand to Save Texflow Image so the round trip closes.

So the minimal graph is: Load Texflow Depth Image → KSampler (img2img on the depth, or a ControlNet pass) → Save Texflow Image, with the TEXFLOW_METADATA output wired straight across. Every pixel you're generating is meant to come back and wrap around the 3D surface.

Where the files come from

The node reads exactly two files, by name, from ComfyUI's input folder: texflow_depth_image.tiff and texflow_occupancy_image.png. The Blender extension uploads them to the same server via the upload/image endpoint - which is why the texflow panel needs the correct ComfyUI URL (usually http://127.0.0.1:8188). Don't rename the files; the names are the contract, fixed on both sides. The node also hashes both files plus the render_id on every check, so when Blender renders a fresh frame, ComfyUI notices and re-runs your graph instead of feeding you a stale depth map.

Install

ComfyUI Manager (search "ComfyUI-texflow-extension"), or the manual route:

cd ComfyUI/custom_nodes
git clone https://github.com/theAdamColton/ComfyUI-texflow-extension

Restart ComfyUI and look for the new "texflow" category. That's the whole install - no pip dependencies beyond what ComfyUI already ships, no models to download, no API keys. The heavy lifting lives on the Blender side, which you install separately as a .zip from the texflow releases page.

Gotchas

• If the node throws a file-not-found, it means Blender hasn't rendered yet - or it's pointed at a different ComfyUI server than the one you're running. Render the depth first, then check the URLs match.
• The render_id lives in the occupancy PNG, not the depth TIFF. Only matters when you're debugging the handshake, but it'll save you an hour if it ever comes up.
• The depth is 16-bit and gets normalized to 0–1, so don't feed a uint8 depth in and expect the same range out.
• The texflow docs recommend an orthographic camera for a flatter, more usable depth map - a genuinely useful tip if your first results come out warped.

It's a quiet little node and you won't find much community chatter around it - the real documentation lives in the texflow README. Treat it as a stable, boring contract: it exists to make the Blender round trip work, and it does.